The effectiveness of this approach was confirmed by comparing the DGI gene scores corrected with step-wise regression analysis to the corresponding gene scores corrected with traditional permutation analysis, as the latter corrects for all confounding effects (Figure 2; see Materials and Methods). The high correlation between the regression-corrected gene scores, and the permutation-corrected gene scores, for all genes (Pearson's correlation coefficients, r = 0.95; p<1e-30, Figure 2B) compared to before correction (r = 0.69, Figure 2A) indicates that only a small fraction of the confounding effects on is not explained by our regression method. Similar results were obtained when gene score ranks were compared (r = 0.95 versus r = 0.82; p<1e-30). A comparison of the distributions of for different sized genes using the permuted DGI data sets, demonstrates that the regression-based correction has indeed removed the confounding effect of gene size on (Figure S2).
